The Muse Ultra deep field (MUDF). II. Survey design and the gaseous properties of galaxy groups at 0.5 < z < 1.5

The MUSE Ultra Deep Field (MUDF). II. Survey design and the gaseous

properties of galaxy groups at 0.5

< z < 1.5

M. Fossati ,

M. Fumagalli ,

E. K. Lofthouse,

V. D’Odorico,

E. Lusso ,

S. Cantalupo ,

R. J. Cooke ,

S. Cristiani,

F. Haardt,

S. L. Morris,

C. Peroux,

L. J. Prichard,

M. Rafelski,

I. Smail

and T. Theuns

1_{Institute for Computational Cosmology, Durham University, South Road, Durham DH1 3LE, UK} 2_{Centre for Extragalactic Astronomy, Durham University, South Road, Durham DH1 3LE, UK}

3_{Dipartimento di Fisica G. Occhialini, Universit`a degli Studi di Milano-Bicocca, Piazza della Scienza 3, I-20126 Milano, Italy} 4_{INAF, Osservatorio Astronomico di Trieste, Via G. B. Tiepolo 11, I-34143 Trieste, Italy}

5_{Scuola Normale Superiore, Piazza dei Cavalieri 7, I-56126 Pisa, Italy}

6_{Dipartimento di Fisica e Astronomia, Universit`a di Firenze, via G. Sansone 1, Sesto Fiorentino, I-50019 Firenze, Italy} 7_{INAF–Osservatorio Astrofisico di Arcetri, I-50125 Firenze, Italy}

8_{Department of Physics, ETH Zurich, Wolgang-Pauli-Strasse 27, CH-8093 Zurich, Switzerland} 9_{DiSAT, Universit`a degli Studi dell’Insubria, Via Valleggio 11, I-22100 Como, Italy}

10_{INFN, Sezione di Milano-Bicocca, Piazza della Scienza 3, I-20126 Milano, Italy}

11_{Centre for Advanced Instrumentation, Durham University, South Road, Durham DH1 3LE, UK}

12_{European Southern Observatory (ESO), Karl-Schwarzschild-Str. 2, D-85748 Garching b. M¨unchen, Germany}

13_{Aix Marseille Universit´e, CNRS, LAM (Laboratoire d’Astrophysique de Marseille) UMR 7326, F-13388 Marseille, France} 14_{Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA}

15_{Department of Physics and Astronomy, Johns Hopkins University, Baltimore, MD 21218, USA}

Accepted 2019 September 17. Received 2019 September 14; in original form 2019 July 30

A B S T R A C T

We present the goals, design, and first results of the MUSE Ultra Deep Field (MUDF) survey, a large programme using the Multi Unit Spectroscopic Explorer (MUSE) instrument at the ESO Very Large Telescope. The MUDF survey is collecting≈150 h on-source of integral field optical spectroscopy in a 1.5× 1.2 arcmin2_{region which hosts several astrophysical structures}

along the line of sight, including two bright z≈ 3.2 quasars with close separation (≈500 kpc). Following the description of the data reduction procedures, we present the analysis of the galaxy environment and gaseous properties of seven groups detected at redshifts 0.5 < z < 1.5, spanning a large dynamic range in halo mass, log(Mh/M_)≈ 11 − 13.5. For four of the

groups, we find associated MgIIabsorbers tracing cool gas in high-resolution spectroscopy of the two quasars, including one case of correlated absorption in both sightlines at distance ≈480 kpc. The absorption strength associated with the groups is higher than what has been reported for more isolated galaxies of comparable mass and impact parameters. We do not find evidence for widespread cool gas giving rise to strong absorption within these groups. Combining these results with the distribution of neutral and ionized gas seen in emission in lower redshift groups, we conclude that gravitational interactions in the group environment strip gas from the galaxy haloes into the intragroup medium, boosting the cross-section of cool gas and leading to the high fraction of strong MgIIabsorbers that we detect.

Key words: galaxies: evolution – galaxies: groups: general – galaxies: high-redshift – galaxies: haloes – quasars: absorption lines.

1 I N T R O D U C T I O N

In the last few decades, observations from ground- and space-based telescopes coupled to theoretical models have built the Lambda

_{E-mail:matteo.fossati@durham.ac.uk}

cold dark matter (CDM) model. In this framework galaxies form within dark matter haloes that grow hierarchically (e.g. Gunn &

Gott1972; White & Rees1978; Perlmutter et al.1999). Several,

and often competing, processes then lead to the growth, evolution, and final fate of galaxies inside these haloes, giving rise to the diverse morphology, colours, and structural properties of present-day galaxies. One of the major tasks of modern theories of galaxy

2019 The Author(s)

formation is thus to describe in detail the physical processes that underpin this observed evolution.

Galaxies grow both in mass and size (Muzzin et al.2013; van

der Wel et al.2014) by acquiring gas either through cooling of a

hot gas halo, or via cold gas streams fed by cosmic filaments (e.g.

Kereˇs et al.2005; Dekel & Birnboim2006; van de Voort et al.

2011). The process of star formation converts the gas into stars

and it is tightly related to the net balance of gas accretion and

ejection (Bouch´e et al.2010; Dav´e, Finlator & Oppenheimer2012;

Lilly et al.2013; Sharma & Theuns2019). Feedback processes

related to stellar winds, supernova explosions, and active galactic nuclei eject gas back into the haloes surrounding galaxies: the circumgalactic medium (CGM). While some of this gas falls back

on the galaxy in a fountain-like mode (Fraternali & Binney2008), a

fraction of it can leave the halo of galaxies contributing to the metal enrichment of the intergalactic medium (IGM; e.g. Dekel & Silk

1986; Schaye et al.2003; Springel et al.2005; Oppenheimer et al.

2010). Within this framework, it becomes crucial for an effective

theory of galaxy evolution to understand how inflows and outflows

interact and coexist within the CGM (e.g. Steidel et al. 2010;

Tumlinson, Peeples & Werk2017), that is the gaseous component

surrounding galaxies.

At the same time, galaxies do not evolve in isolation, but are accreted on to more massive structures, comprising groups or clusters of galaxies. Indeed, it has long been known that galaxies in dense environments have different properties than those in less dense environments, both in terms of gas content and structural

properties (e.g. Oemler1974; Dressler1980; Giovanelli & Haynes

1985; Balogh et al.2004; Peng et al.2010; Fossati et al. 2017).

Environmental processes, triggered by the interactions between galaxies themselves with the hot medium in massive haloes and with the IGM further regulate the gas supply of galaxies (e.g. Boselli &

Gavazzi2006), leading to a different evolution compared to more

isolated galaxies.

Significant progress has been made in understanding the gas– galaxy coevolution since the advent of large spectroscopic surveys

of galaxies at z 1 (e.g. SDSS; York et al.2000or GAMA; Driver

et al. 2011). These galaxy surveys have been complemented by

spectroscopic observations of quasars allowing for detailed studies of the CGM in absorption as a function of galaxy properties (including mass, star-formation rates), and their environment (e.g.

Prochaska et al.2011; Stocke et al.2013; Tumlinson et al.2013;

Tejos et al.2014; Bordoloi et al.2014b; Finn et al.2016; Kauffmann

et al. 2017; Tumlinson et al. 2017). These studies reveal the

ubiquitous presence of a multiphase, enriched, and kinematically

complex CGM surrounding every galaxy (Werk et al.2014,2016).

This picture of a multiphase CGM has been extended to z > 1 thanks to extensive observational campaigns with multi-object

spectrographs on 8–10 m telescopes (e.g. Rubin et al.2010; Steidel

et al.2010; Crighton et al.2011; Rudie et al.2012; Tummuangpak

et al.2014; Turner et al.2014; Turner et al.2017). Despite these

advancements, however, our view of the CGM at early cosmic epochs has been mostly limited to star-forming galaxies at the bright

end of the UV luminosity function, and to scales of≈ 0.1–1 Mpc

around galaxies, due to the difficulty in obtaining spectroscopy of samples of objects at small projected separations from the quasars using traditional multi-objects spectrographs.

These limitations have recently been lifted by integral field spectrographs which have been deployed at the largest observing facilities, including the Multi-Unit Spectroscopic Explorer (MUSE;

Bacon et al.2010) at the ESO Very Large Telescope, and the Keck

Cosmic Web Imager (KCWI; Morrissey et al.2018) at the Keck

Observatory. In particular, thanks to its large 1 arcmin2_{field of view,}

its extended wavelength coverage in the optical, and its exquisite sensitivity, MUSE has become the ideal instrument for studies of

the CGM of galaxies in quasar fields (e.g. Fumagalli et al.2016,

2017b; Schroetter et al.2016; Bielby et al.2017; P´eroux et al.2017;

Klitsch et al.2018; Chen et al.2019).

Leveraging the unique features of this instrument, we have designed an observational campaign to acquire very deep MUSE

observations in the field centred at 21h_:42m_:24s_{, −44}◦_:19m_:48s

(hereafter the MUSE Ultra Deep Field or MUDF). This field stands

out for its rare property of hosting two bright quasars at z ≈

3.22 that probe the IGM and CGM of intervening galaxies with

two sightlines≈60 arcsec apart. Another quasar lies at the same

redshift at≈8 arcmin separation, making this system a quasar triplet

(Francis & Hewett1993). Upon its completion, this programme will

acquire≈200 h of MUSE data (corresponding to ≈150 h on source)

in a 1.5× 1.2 arcmin2_{region around the two quasars (ESO PID}

1100.A−0528). This programme is complemented by deep high-resolution spectroscopy of the quasars using the UV and Visual

Echelle Spectrograph (UVES; Dekker et al.2000) at the VLT (ESO

PIDs 65.O−0299, 68.A−0216, 69.A−0204, 102.A−0194), and by

the deepest spectroscopic survey (90 orbits in a single field) in the near-infrared using the Wide Field Camera 3 instrument on board the Hubble Space Telescope, together with deep eight-orbit near UV imaging (HST PIDs 15637 and 15968).

These combined data sets will enable us to achieve several goals. First and foremost, we will connect the presence of gas in the CGM

of galaxies with their properties and their environment from z≈ 3 to

the present day. Without a pre-selection for UV and optically bright sources, we will have a unique vantage point on the low mass galaxy

population up to z ∼ 3. Furthermore, the MUDF hosts notable

structures as a function of redshift. For instance, a correlated strong

HIabsorber detected in both sightlines at z≈ 3 hints at an extended

structure running across the field of view (D’Odorico, Petitjean &

Cristiani2002). Moreover, the presence of a quasar pair is suggestive

of an overdense region at z≈ 3.22, which is predicted to lie at the

intersection of filaments in the cosmic web. Indeed, in the first paper

of this series, Lusso et al. (2019) studied the morphology of the

giant Ly α nebulae surrounding the quasars, finding an elongation of the ionized gas along the line connecting the two quasars. In the future, once we have the full data set, we will search for the presence of ionized gas in this putative filament. The depth of the observations will also provide spectra of exquisite quality for a few hundred objects. Thanks to our multiwavelength data set from the near-UV to the near-IR, we will study the properties and structure of low-mass galaxies across a large fraction of cosmic time.

In this paper, we present the survey design and the details of the MUSE observations and data reduction. As a first application

we focus on the connection of enriched cool gas (T ∼ 104_K),

as traced by the MgIIλλ2796, 2803 Å absorption doublet in the

quasar spectra, with the galaxy population and its environment at 0.5 < z < 1.5. Even though the full MUSE data set is still being collected, the observations available to date already provide an excellent data set for an accurate reconstruction of the local galaxy

environment, and of the physical properties of galaxies at z 1.5.

Several studies have used the MgIIdoublet to trace gas with similar

column densities to that detected through 21-cm atomic hydrogen

observations (e.g. Bergeron 1986; Steidel, Dickinson & Persson

1994; Chen & Tinker2008; Kacprzak et al.2008; Chen et al.2010;

Gauthier2013; Bordoloi et al.2014a; Nielsen et al.2015; Schroetter

et al.2016; Nielsen et al.2018; Rubin et al.2018a,b). These studies

have found that MgII absorbers trace the CGM of galaxies and

possibly outflowing gas up to a distance of∼100 kpc (Kacprzak

et al.2008; Chen et al.2010). This transition is therefore ideal to

study the CGM of galaxies in groups and in isolation in the MUDF, for the first time in a very deep and complete data set.

The paper is structured as follows: first we present the MUDF survey strategy, the data reduction procedures, and the quality validation of the MUSE data (Section 2), and of the high-resolution UVES spectroscopy (Section 3). We then describe the procedures adopted to extract the sources and their properties (Section 4), and the reconstruction of the local environment by searching for groups of galaxies in the field (Section 5). In Section 6, we describe a novel method to fit the high resolution quasar spectra to extract metal absorption profiles and we present our results on the correlation of absorbers and galaxies in groups and in isolation. We conclude with a discussion of these results (Section 7) and with a summary of our findings (Section 8).

Throughout this paper, we assume a flat CDM cosmology with

H0= 67.7 km s−1Mpc−1and m = 0.307 (Planck Collaboration

XIII 2016). All magnitudes are expressed in the AB system,

distances are in proper units, and we assume a Chabrier (2003)

stellar initial mass function.

2 M U S E O B S E RVAT I O N S 2.1 Survey strategy and current status

The science goals of the MUDF programme include the study of the galaxy population around and along the line of sight to the quasar pair, a deep search for Ly α emission from the putative filaments

which are expected to connect the two quasars at z≈ 3.22. Because

the projected distance of these quasars is≈62 arcsec on the sky, a

single Wide Field MUSE pointing of 60 arcsec on a side (the exact shape is trapezoidal) would not allow a full mapping of the area of interest. For these reasons, we designed an observational strategy that includes two heavily overlapping pointings: North–West and South–East (hereafter named simply North and South), the centres

of which are shown as black crosses in Fig.1. Throughout the

entire survey, we plan to collect∼200 frames dithered around each

of these centres. The nominal exposure time of each frame is 1450 s and different frames include small on-sky dithers (≈3–4 arcsec), as well as 10 deg rotations of the instrument to reduce systematic errors arising from the different response of the 24 spectrographs and detectors of MUSE. While multiple exposures will be taken at the same rotation angle, the sequence of dithers has been designed to ensure that those exposures will not have the same centre and orientation.

The final survey footprint will be ellipsoidal, with a≈−45 deg

position angle (North through East), and a major and minor axis of ≈110 and 90 arcsec respectively. At the time of writing, we have

reached∼ 35 per cent completion, and Fig.1shows the exposure

map generated with these data overlaid on a combination of white-light images from the MUSE data itself and from the Dark Energy

Survey (DES) Data Release 1 (Abbott et al.2018) outside the MUSE

footprint. The survey design prioritizes the collection of deeper data in the area between the two quasars, where our science goals warrant maximum sensitivity. The outer black dashed contours mark the regions with at least 15 and 30 h of exposure, respectively.

The instrument set-up makes use of the MUSE Wide Field Mode

with extended wavelength coverage in the blue (4650−9300 Å),

in order to search for Ly α emitting galaxies down to z≈ 3. We

also take advantage of the Ground Layer Adaptive Optics module (GALACSI) which uses artificial laser guide stars to improve the

Figure 1. Exposure map (colour scale) of the MUSE observations in the MUDF overlaid on an optical white light image from the MUSE data where available or from the Dark Energy Survey combined g, r, i images elsewhere (shown in grey on the same surface brightness scale). North is up and East is left. The outer and inner black dashed contours mark the regions with at least 15 and 30 h of exposure, respectively, in this partial data set. The black and the white crosses mark the pointing centres and the position of the two quasars, respectively.

image quality by partially correcting for atmospheric turbulence. In this way, the observations can be performed under a wider range of natural seeing conditions without compromising the image quality of the final mosaics. However, this set-up implies that we cannot use data between 5760 and 6010 Å, a range affected by the sodium line generated by the laser beam. Moreover, laser induced Raman scattering of molecules in the atmosphere creates emission lines outside this range. These lines are removed as part of the sky subtraction data reduction steps with no impact on the data quality. In this paper, we present results obtained with the first 44.1 h of observations. The data are collected as part of ESO programme 1100.A-0528. The first run (A) was executed in visitor mode on 2017 August 16, where we acquired 19 exposures of 1270 s each in dark-time. The conditions were excellent with clear skies and an image quality of 0.4–0.6 arcsec in the final reconstructed frames. After this successful run, further dark-time observations have been executed in service mode (runs B and C) under similarly good observing conditions. As of 2019 June, we have acquired 93 frames with an exposure time of 1450 s each. In total we have acquired 159 ks of data, corresponding to 44.1 h. The image quality on the

final coadd has a full width at half-maximum, FWHM= 0.57 arcsec

as measured by Moffat profile fits to point sources.

2.2 Data reduction

The data reduction procedure follows the methodology described

in previous works by our team (Fumagalli et al. 2016, 2017a;

Lofthouse et al.2019). In brief, we use the ESO pipeline (v2.4.1;

Weilbacher & et al.2014) to reduce the calibrations (bias, flat, arcs,

and standard stars) and apply them to the individual exposures. We

also use this software to resample the detector values into cubes that are then sky subtracted using models of the sky continuum and sky lines that are matched to the data extracted from the darkest pixels in each frame. The individual exposures are then aligned by using the position of point sources in the field, and we generate a first coadd of the frames. We register this initial coadd to match the WCS coordinates of the two quasars from the Data Release 2

of the GAIA survey (Gaia Collaboration2018). The combination of

relative offsets (arising from dithers) and this absolute WCS offset (arising from the pointing accuracy of MUSE) is then propagated into the pixel-tables of each exposure. We then reconstruct the cubes again for each exposure, this time on a pre-defined spatial grid of

540× 540 pixels, with each voxel (volumetric pixel) measuring

0.2 arcsec in the spatial direction and 1.25 Å in the spectral direction. The size of the grid has been chosen such that all the spatially aligned exposures of the programme will fall on to it, which eliminates the need for re-projecting the data when new observations are acquired. The cubes produced by the ESO pipeline are not yet fully science-grade, as they are still affected by an uneven spatial illumination

and by residuals from sky lines subtraction (Bacon et al.2017). To

correct for these effects, we post-process the individually aligned, reconstructed, and un-skysubtracted exposures using routines from

the CUBEXTRACTOR package (v1.8,CUBEXhereafter, Cantalupo

in preparation; see Cantalupo et al.2019for a description of the

algorithms). The adopted procedure follows earlier work using

MUSE data (Borisova et al.2016; Fumagalli et al.2016,2017a).

In brief, we use theCUBEFIX tool to correct residual differences

in the relative response of the 24 MUSE IFUs and of individual slices, which are not fully corrected by flat-field calibration frames.

CUBEFIX improves the illumination uniformity by measuring the average illumination in each stack (the MUSE FoV is composed of 24 IFUs which are further made of 4 stacks of 12 slices), as a function of wavelength on white-light images generated on-the-fly

from the cube. We then use theCUBESHARP tool for sky subtraction.

The algorithm performs a local sky subtraction, including empirical corrections of the sky line spread function. The combination of these two routines is applied twice, by using the first illumination-corrected and sky-subtracted cube to mask the sources in the field for the second iteration. The masking step is critical to measure the true instrumental illumination in each stack and to achieve a high-quality sky subtraction.

After this first double pass of theCUBEX tools on the individual

frames, we coadd them with mean statistics and 3σ clipping to generate a deep white light cube. This cube is then used to detect

and mask the sources and the deep mask is then fed intoCUBEFIX

andCUBESHARP for a final run. After this step we measure the FWHM of point sources, their average flux, and their position with respect to the GAIA astrometry using Moffat fits to the white light images of each exposure. The distribution of these values across all the exposures are then inspected to ensure that all the frames are correctly aligned and flux calibrated. The photometry is consistent at a 4 per cent rms level across frames and the astrometric precision is within 0.05 and 0.03 arcsec in RA and Dec. respectively, i.e.

<10 per cent of the spatial resolution. Moreover, no frame has

been identified to be a >3σ outlier on any of these metrics and therefore we include all frames in the final combine.

Lastly, we combine the individual cubes with a 3σ clipping rejection of outliers and both with mean and median statistics. We also generate mean combines obtained with two independent halves of the exposures. These products are useful to correctly identify weak emission-line sources in the cube from residual artefacts (e.g. residuals of cosmic-ray hits), which are likely to appear only in one

Figure 2. Histograms of flux in voxels values, normalized by the voxel standard deviation from the ESO+CUBEXpipeline in three wavelength ranges (blue, orange, and green), and in the full wavelength range after reconstructing the variance cube using a bootstrap resampling technique (red). Shown with black lines are Gaussian fits to these distributions, with the resulting standard deviation listed in the legend. Only spatial pixels free from continuum sources are included in this analysis.

of the two independent combines. TheCUBEX reduction pipeline

significantly improves the quality of the illumination uniformity across the entire observed area. We quantified the flatness of the illumination by comparing the standard deviation of the flux in sky

pixels of the white-light image from theCUBEX processing relative

to the ESO reduction, finding a ratio of 0.23. This means that the products used in this work are four times deeper than what would have been possible to achieve with the standard pipeline.

2.3 Noise characterization

During each step of the reduction process, the Poisson noise from detector counts is propagated and then combined into a cube that contains the variance of the resampled flux values. However, during several steps, including the drizzle interpolation on to the final grid, the propagated variance departs from accurately reproducing the effective standard deviation of individual voxels in the final data

cube (see Lofthouse et al.2019). In Fig.2we show the flux

distri-bution of voxels (fvox) normalized by the pipeline error (σ1) in each

pixel within three wavelength intervals that are increasingly affected by atmospheric sky lines (namely 4900–5500 Å, 6400–7000 Å, and 7800–8400 Å, in blue, orange, and green, respectively). Once sources are masked, the distribution is expected to approximate a Gaussian (the black lines are Gaussian fits to the distributions), with

standard deviation of unity. Instead, Fig.2shows that the pipeline

error underestimates the true standard deviation in regions free from

sky lines (blue histogram, with σ = 1.19), while it overestimates

it in regions more contaminated by skylines (green histogram, with

σ = 0.83). Moreover, the distribution of fvox/σ1in the wavelength

interval 7800− 8400 Å shows the largest departure from a Gaussian

distribution, an effect that could be attributed to the second-order contamination of the spectra due to our use of the MUSE extended wavelength mode.

We overcome this issue by bootstrapping the combination of individual exposures for each pixel to accurately reconstruct the noise in the final mean, median, and half-exposure cubes. For this, we use 10 000 realizations of the bootstrap procedure to produce a variance cube. We then inspect the flux distribution of pixel

Figure 3. Ratio between the flux dispersion in apertures of varying size (σeff) and the error derived propagating the variance (σN), as computed

across the MUDF datacube in a 5 Å window (4 pixels) as a function of wavelength and aperture size. The dark stripe just below 6000 Å is the region without valid data due to the scattered light from the AO laser.

values divided by the standard deviation of this new error, finding a distribution much closer to Gaussian, although still offset by a few per cent from unity. We attribute this offset to a non-Gaussian

distribution of the fvoxvalues. Due to this small offset, we further

rescale the bootstrap variance cube to obtain a distribution of fvox/σ1

that is unity on average, as shown by the red line in Fig. 2.

We also note that by adopting this improved variance cube, the trend with wavelength which affects the propagated variance is also removed.

Lastly, we derive a model for the correlated noise arising from the resampling of the pixel tables on to a final grid, as described

in Lofthouse et al. (2019). This model represents the correction

that needs to be applied to the propagated error for a source in a

square aperture of N spatial pixels on a side, σN, to recover the

effective noise, σeff. In the spectral direction we use an aperture of

4 pixels (≈ 5 Å) which is generally appropriate for narrow emission

lines in galaxies. Fig.3shows this correction in bins of wavelength

and aperture size. While the dependence with aperture size is a smooth and monotonic function, the wavelength trend is not trivial to understand. It does not correlate to the brightness of sky lines, and we hypothesize it might be driven by the optomechanical design of the instrument, by our observing strategy, or more likely by a combination of the two. Regardless of the physical origin, when using these data to search for line emitters, we will use a

second-order polynomial fit that describes σeff/σN as a function of the

aperture size, for the wavelength bin where each source is found.

3 H I G H - R E S O L U T I O N Q UA S A R S P E C T R O S C O P Y

As part of the MUDF survey, we are collecting a set of ancillary data to complement the MUSE observations. In this work, we make use of UVES high-resolution spectroscopy of the two quasars,

J214225.78-442018.3 (also known as Q2139− 4434, and hereafter

QSO-SE) at z= 3.221 ± 0.004 and J214222.17-441929.8 (Q2139 −

4433, hereafter QSO-NW) at z= 3.229 ± 0.003 (Lusso et al.2019),

for which we provide a description of the data acquisition and data reduction.

Some UVES data for the two quasars already exists in the ESO archive (PIDs 65.O-0299, 68.A-0216, 69.A-0204; D’Odorico et al.

2002). These data cover the wavelength range≈4100–9000 Å, with

a gap between ≈7400 and 7500 Å, due to the gap between the

two CCDs in the red arm of UVES. The S/N of the bright quasar

is≈25 per pixel across most of the wavelength range, while the

fainter quasar has a S/N≈8 per pixel. To increase the wavelength

coverage (down to ≈3600 Å and to fill the current gaps) and to

increase the S/N of the fainter quasar, we have been awarded a total of 23 h of new UVES observations (PID 102.A−0914). At the time of writing, observations for QSO-SE have been completed for a total of 7 h on-source, while for QSO-NW only 15.5 h were obtained, and 17 h are still to be observed. In this paper, we therefore make use of the full data set for the brighter quasar, relying only on a partial data set for the fainter one. We will describe the spectra obtained from the complete observations in a forthcoming paper.

All data were reduced with the current version of the UVES pipeline (v. 5.10.4), using default parameters and procedures. At the end of the standard reduction process, the merged, non-rebinned spectra were reformatted with a custom script and input to

the ESPRESSO Data Analysis Software (DAS; Cupani et al.2016)

for the final operations of coaddition and continuum fitting. This step avoids multiple rebinning of the spectra, which would introduce correlations in the error array. Spectra have been normalized to the continuum estimated by the ESPRESSO DAS. This software fits a cubic spline to the spectrum redwards of the Ly α emission. In the Ly α forest, the fit is iteratively improved by the simultaneous fit of the Ly α absorption lines (for more details see Cupani et al.

2016). The final continuum-normalized spectra were then rebinned

to a constant velocity step of 2.5 km s−1.

4 T H E G A L A X Y P O P U L AT I O N I N T H E M U D F In this section we characterize the galaxy population detected in the MUDF and describe the procedures used for the identification of continuum sources, the extraction of their spectroscopic redshifts, and the derivation of their physical properties through stellar population synthesis fits.

4.1 Source detection

We identify continuum sources using theSEXTRACTOR(Bertin &

Arnouts 1996) software on the white light image reconstructed

from the MUSE datacubes. We input a variance image and we use a conservative threshold of 3σ above the local noise, and a minimum area of 10 pixels for detection. The minimum deblending parameter DEBLEND CONT is set to 0.0001, chosen to enable the detection of sources in crowded regions of the mosaic. We restrict source extraction to the area where we collected more than 10

exposures, corresponding to an observing time of≈4 h, to avoid

spurious detections at the noisy edges of the field of view. In future publications we will use deep HST imaging for source detection in the field.

This procedure identified 250 sources. For each of them, we

extract the magnitude in the detection image (mMUSE) in an elliptical

aperture with size equal to 2.5 times the Kron (1980) size from

SEXTRACTOR. We also reconstruct a 1D spectrum summing the spectra from pixels within the Kron aperture, also transforming the wavelength to vacuum. In both these procedures, we mask nearby sources whose segmentation map falls in the extraction mask to minimize the effects of blending.

Table 1. The properties of the continuum sources in the MUDF extracted bySEXTRACTORwith S/N > 3. Column 1 shows the source ID, column 2 shows the source name. Columns 3 and 4 list the right ascension and declination of the sources followed by the MUSE white-light magnitude of the source in column 5 with its associated error (column 6). The redshifts obtained using MARZ are shown in column 7 followed by their confidence (column 8). A confidence flag of 3 or 4 indicates reliable redshifts while flags 1 or 2 are for unknown or highly uncertain redshifts, respectively. Column 9 shows the stellar mass from stellar population fitting for sources with z < 1.5. The full table is included as online only material.

ID Name RA (J2000) Dec. (J2000) mMUSE σmMUSE Redshift Confidence log (M∗/M)

1 MUDFJ214226.11-442031.3 325.60878 −44.34204 24.3 0.2 4.0441 4 –

2 MUDFJ214223.99-442029.5 325.59996 −44.34154 26.0 0.3 – 1 –

3 MUDFJ214224.73-442025.4 325.60304 −44.34039 25.7 0.3 – 1 –

4 MUDFJ214223.34-442029.4 325.59725 −44.34151 24.3 0.1 – 1 –

5 MUDFJ214225.93-442026.3 325.60805 −44.34064 23.6 0.1 1.0530 4 10.24

Figure 4. Reconstructed optical white-light image from the MUSE data (where available) or from the DES g, r, i images (outside the MUSE FoV). North is up and East is left. The two quasars are marked with white crosses. Continuum-detected sources identified within the MUSE field are marked with apertures (red for sources with spectroscopic redshifts with confidence classes 4 and 3) with size equal to six times the Kron radius. The dotted orange contour encloses the region where the exposure time is≥4 h.

We measure redshifts using the MARZsoftware (Hinton et al.

2016), which we customize1_{with the inclusion of high-resolution}

synthetic templates for passive and star-forming galaxies at z

< 2 derived from Bruzual & Charlot (2003) stellar population

synthesis models. Following automatic template fitting, individual sources are inspected and classified by two authors (MFo and EKL) in four classes (4, secure redshift with multiple absorption or emission features; 3, good redshift with single but unambiguous feature; 2, possible redshift, based on a single feature; 1, unknown

1_{This version is available athttps://matteofox.github.io/Marz}

redshift). Typical redshift uncertanties are δz≈ 0.0002 × (1 + z),

corresponding to δv≈ 60 km s−1.

The final redshift classification is presented in Table1. Fig.4

shows the position within the MUDF footprint of the

continuum-detected sources and theirSEXTRACTORIDs. Hereafter we make use

only of the objects with a reliable spectroscopic redshift (classes 3 and 4) which are marked with red apertures. The redshift distribution of these 117 sources (47 per cent of the detected sample) is shown in

Fig.5. The range 1.5 < z < 2.5, known as the ‘redshift desert’, does

not have significant detections due to the absence of strong emission lines that would fall within MUSE coverage. Only two galaxies

are identified there thanks to strong AlIIIabsorption lines in their

Figure 5. Distribution of spectroscopic redshifts with confidence classes 4 and 3 (see the text for details) for continuum sources in the MUDF.

spectra. This range, however, will be filled in by the ultra-deep near-infrared observations which we are collecting with HST/WFC3.

4.2 Completeness

To assess the completeness of our source extraction procedure, we inject mock sources of known magnitude into the detection (white-light) image and we assess their detection rate. First, we inject 2D Gaussian templates with a FWHM of 0.6 arcsec to simulate unresolved point sources at the resolution limit of the final MUSE mosaic. Then, we repeat the experiment using circular exponential profiles which are more appropriate for real disc-like galaxies. In this case we use an exponential scale length of 0.26 arcsec which

corresponds to an effective radius of 5 kpc at z∼ 1, which is typical

for star-forming galaxies (van der Wel et al.2014). The intrinsic

exponential profiles are convolved with a Gaussian kernel with a FWHM of 0.6 arcsec to account for the effects of the observational PSF, and do not include the effects of the disc inclination. In each iteration, we inject 80 mock sources (to avoid confusion and blending issues) in blank background regions and we repeat the detection procedure 10 000 times.

The bottom panel of Fig.6shows the fraction of detected mock

sources as a function of magnitude for sources injected at locations of the image with different exposure times. For point sources (solid lines) we reach fainter limits than for extended sources (dashed lines), due to their compactness. The arrows mark the faintest magnitude at which we are 90 per cent complete in each bin of exposure time. In the deepest bin, where we have so far

collected between 30 and 45 h of data, we reach mMUSE 27.8

and 26.6 mag for point-like and extended sources, respectively. The

top panel of Fig.6 shows the number of detected galaxies as a

function of the white-light magnitude, mMUSE. We fit a linear relation

(dashed line) to the logarithm of the number counts in the region

24 mag < mMUSE<26.5 mag, where our survey is complete. We

then show in the bottom panel the empirical detection fraction (black points) determined as the ratio of the detected galaxies relative to the expected number from the linear fit. Despite the low number statistics, we find that this empirical detection fraction is in between the limits defined by the two mock experiments (point sources and extended sources), complementing and corroborating the approach based on mocks.

Figure 6. Top panel: Number of galaxies extracted as a function of the magnitude in the detection image. The grey dashed line shows the expected number of galaxies determined from a linear fit to the region between mMUSE>24 mag and mMUSE<26.5 mag. Bottom panel: Fraction of mock

sources detected bySEXTRACTOR. The solid lines are for the injection of Gaussian sources with a FWHM of 0.6 arcsec. The dashed lines are for sources with an exponential profile with scale-length equal to 0.26 arcsec and PSF convolved. Lines of different colours are for different exposure times in the MUDF map. The arrows (filled for point sources and empty for extended ones) mark the faintest magnitude at which we are 90 per cent complete in each bin of exposure time. The black points (with error bars) show the empirical detection fraction determined as the ratio of the detected galaxies relative to the expected number from the linear fit in the top panel.

Table 2. Name and wavelength range of the top-hat filters defined for the photometric extraction.

Filter Name λmin(Å) λmax(Å)

MUSE Blue 4750 5700

MUSE Green 6100 7100

MUSE Red 7100 8200

MUSE NIR 8200 9300

Lastly we note that, to reach maximum depth, the detection image is obtained from the full MUSE wavelength range, which does not correspond to the most commonly used broad-band filters. To facilitate the comparison to other imaging surveys we have computed the following colour corrections for a star-forming

(passive) galaxy template at z= 1: mMUSE− rSDSS= −0.31(−0.44),

and mMUSE− iSDSS= 0.04(0.05).

4.3 Source photometry and ancillary data

Taking advantage of the wide wavelength coverage of MUSE, we extract source photometry in four pseudo-filters with the goal of characterizing the physical properties of the galaxy population. The width and the number of filters have been carefully selected to maximize our sensitivity to strong breaks in the galaxy spectra, while keeping a good S/N ratio in the measurements, and avoiding the gap in the spectra due to the AO laser filter. We convolve the MUSE datacube with the top-hat filters, whose ranges are given in

Table2, to generate an image and the associated variance. We then

run a forced photometry algorithm using apertures with radius 2.5×

rKronas defined above for the detection image.

The MUSE data probe the rest-frame near-UV to optical region

of the galaxy spectra for sources at z 1.5. While providing a good

sensitivity of the recent star-formation activity and stellar content of the galaxies, these blue wavelengths are affected by dust extinction which could bias our reconstruction of the galaxy parameters.

To mitigate this issue, we include in our analysis data taken with

the IRAC (Fazio et al.2004) instrument onboard the Spitzer (Werner

et al.2004) Space Telescope. IRAC imaging of an area of the sky

including the MUDF has been taken during the cryogenic mission

(PID: GO-3699; Colbert et al.2011) using the four channels at 3.6,

4.5, 5.8, and 8.0 μm. However, due to the field of view offset of the IRAC camera, only the channels at 3.6, and 5.8 μm cover the MUDF in its full extent, and we use them in this work. The total integration time per pixel of these observations are 1800 s, and image FWHM

is≈1.8 arcsec.

To extract IRAC photometry from these images we use theT

-PHOTsoftware (Merlin et al.2015), which optimally deals with the

lower resolution of IRAC data in crowded fields by means of source priors from higher resolution imaging. We use the MUSE white light image, PSF model, and source catalogue as the high resolution data set. We then derive a Gaussian convolution kernel for each IRAC channel, to transform the MUSE PSF into the IRAC one. Lastly we

resample the IRAC data on to the MUSE pixel grid and runT-PHOT.

Due to the shallowness of the IRAC data, only the relatively bright sources can be detected robustly, especially in crowded areas of the

MUDF. In the end, we visually inspected theT-PHOTresults for each

source to assess if results: (i) can be used and the detection is at least with 3σ significance; (ii) can be used as an upper limit; (iii) cannot be used because the source is faint and it is highly blended with a nearby bright source. In the first two cases (95 per cent of the sources we fit), we include the IRAC data in the estimates of the stellar populations as described in the following section.

4.4 Stellar population parameters

We characterize the physical properties of the galaxy population in the MUDF by jointly fitting the MUSE spectra and the photometric data derived from MUSE and IRAC as described above. These combined data sets allow us to derive reliable estimates for the star-formation history, stellar mass, current star-formation rate, and dust content of the galaxies. This multiwavelength approach is instrumental in breaking the degeneracy that often arise between these parameters, most notably between stellar mass and dust

extinction, or star-formation rate and galaxy age (Wuyts et al.2011;

Fossati et al.2018). In this work, we primarily use the derived

stellar masses, and we postpone the discussion of results involving the other quantities to a follow-up paper, where we will re-assess the quality of these measurements by adding into the fitting procedure the near-UV and near-IR data that will be collected with HST as part of programmes PID 15637 and 15968.

To infer the source properties, we fit the observed data to synthetic models using the Monte Carlo Spectro-Photometric Fitter (MC-SPF). A more complete description of the code is given in

Fossati et al. (2018); here we briefly describe the procedure and

the model set. First, we build a grid of synthetic stellar spectra

from the Bruzual & Charlot (2003) high-resolution models at solar

metallicity. Following similar work that studied the properties of

galaxies in deep fields (Wuyts et al.2011; Momcheva et al.2016),

we use exponentially declining star-formation histories with e-folding times varying between 300 Myr and 15 Gyr, and galaxy ages varying between 50 Myr and the age of the Universe at the redshift of each galaxy. With the imposed minimum e-folding time, Wuyts

et al. (2011) found that the SFR derived with this spectral energy

distribution technique matches the one obtained from UV+

far-IR photometry for systems with low-to-intermediate star-formation

rates (SFR 100 M_yr−1), which are likely to dominate the sources

in our small field.

We further add nebular emission lines to the stellar template,

by using line ratios from Byler et al. (2018) scaled to the line

luminosity using the number of Lyman continuum photons from the stellar spectra. The SFH and nebular emission grids are interpolated on-the-fly during the likelihood sampling. We assume a double

Calzetti et al. (2000) attenuation law to include the effects of dust

attenuation. Stars older than 10 Myr and emission lines arising from

them are attenuated by a free parameter, Aold, while younger stars

are attenuated by Ayoung= 2.27 × Aoldto include the extra extinction

occurring within the star forming regions.

We jointly fit the photometric data points and the MUSE spectra at their native resolution. A third degree multiplicative polynomial is used to normalize the spectra to the photometric data and to remove large-scale shape differences between the models and the spectra, especially at the edges of the wavelength range where the MUSE response function is more uncertain. The multidimensional

likelihood space is sampled by usingPYMULTINEST(Buchner et al.

2014), a python wrapper for theMULTINESTcode (Feroz & Hobson

2008; Feroz et al.2013). Fig.7shows an example of the results

obtained with this fitting procedure for a galaxy in the MUDF. With the current wavelength coverage, some degeneracy remains between dust extinction and star-formation rate. Rest-UV data from HST will break this degeneracy, however, in this work we primarily make use of the stellar mass estimates from the fits which are found to be well converged and free from degeneracies with other parameters.

5 G R O U P I D E N T I F I C AT I O N

The detection of a large number of sources with accurate

spec-troscopic redshifts in narrow redshift bins (see Fig. 5) is highly

suggestive of the presence of overdense structures in the MUDF footprint. Indeed, overdensities spanning from compact groups to clusters and superclusters of galaxies have been detected in all the fields which have been targeted by deep and extensive spectroscopic

campaigns (e.g. Scoville et al. 2007; Yang et al. 2007; Kovaˇc

et al.2010; Diener et al.2013; Balogh et al.2014; Fossati et al.

2017; Galametz et al.2018). We therefore proceed to systematically

identify galaxy groups in the MUDF.

There is a rich collection of literature on finding groups in spectroscopic redshift surveys, with most methods based on a

Friends-of-Friends approach (Huchra & Geller1982; Berlind et al.

2006; Knobel et al.2009,2012; Diener et al.2013). These methods

link galaxies into structures by finding all objects connected within linking lengths r (a physical distance) and v in redshift space. The choice of these parameters is driven by the need to balance the competing requirements of identifying all group members without overmerging different groups, avoiding interlopers, and taking into account redshift uncertainties. In this work we search for galaxy groups at 0.5 < z < 1.5, where the lower limit is dictated by the small volume probed at lower redshift and the upper limit is given by the redshift desert. In this range, we have highly accurate spectroscopic redshifts for all the galaxies that we aim to connect into structures.

We use r= 400 kpc and v = 400 km s−1, following Knobel et al.

(2009) in the same redshift range. Similar to previous works, we

define a galaxy group to be an association of three or more galaxies

(Knobel et al.2012; Diener et al.2013), independently of stellar

mass or observed magnitude.

Figure 7. Example of the stellar population fitting results for source ID 35 in the MUDF. Rest-frame MUSE spectrum (black line) and best fit model (red line). The fit residuals (Data-Model) are given below the main panel, and the shaded area shows the noise array. The inset shows the spectral energy distributions extracted from the MC-SPF posterior (red lines) with the observed photometry overplotted in black. The blue lines show the contribution of young stars (Age <10 Myr) to the total template.

Table 3. Properties of the groups identified in the MUDF. The table lists: the group ID; the average RA, Dec., and redshift of the group members; the number of members in each group; the halo mass derived from the stellar-to-halo mass relation of Moster et al. (2010); the virial radius (R200) for that halo mass; the equivalent width of the MgIIλ2796 Å absorption line associated with the group (or 2σ upper limit) along the QSO-SE and QSO-NW sightlines.

ID <RA> <Dec.> <z > Ngal log (Mhalo/M) rvir W2796 W2796

J2000 J2000 kpc (QSO-SE) Å (QSO-NW) Å 1 325.602433 −44.332283 0.67837 11 12.0 160 0.85+0.03_−0.04 <0.08 2 325.598267 −44.332865 0.68531 5 11.2 87 <0.02 <0.07 3 325.593548 −44.330041 0.78491 3 10.8 61 <0.02 <0.06 4 325.597824 −44.330187 0.88205 6 12.9 293 1.34+0.02_−0.01 0.43+0.01_−0.01 5 325.604431 −44.331396 1.05259 15 13.4 419 1.67+0.02_−0.02 <0.06 6 325.605063 −44.331111 1.15524 3 12.1 138 1.16+0.06_−0.13 <0.06 7 325.602279 −44.328677 1.22849 4 11.6 96 <0.02 <0.06

Following this procedure, we find seven groups in the MUDF

footprint, and their properties are listed in Table3. We estimate the

group halo mass by summing the stellar mass of the group galaxies

(see Yang et al.2007, for a validation of the method) and using the

stellar to halo mass relation from Moster et al. (2010) at the redshift

of the group. We also report in the table the geometric centre of the group, the average redshift of its members (not weighted by other galaxy parameters), and the virial radius calculated from the halo mass. Assuming a typical uncertainty on the total stellar mass

of 0.15 dex (Conroy, Gunn & White2009; Gallazzi & Bell2009;

Mendel et al. 2014), this turns into an error on the virial mass

of 0.20–0.25 dex depending on the local gradient of the stellar-to halo-mass relation. This uncertainty corresponds stellar-to an error of ∼50−60 kpc on the virial radius. The virial radius is further affected by the implicit assumption of virialization of the group halo. With only a few members per group, this cannot be guaranteed as we might be observing groups in formation. Given these caveats, the estimated virial radii should only be taken as indicative values, especially for structures with less than 10 members.

At the redshift of the groups, we search for strong line emitters

which are too faint in continuum to be detected. We runCUBEXon

the continuum-subtracted cube and we use the detection parameters

defined in Lofthouse et al. (2019), i.e. voxel S/N > 3, minimum

number of voxels of 27 and minimum number of channels in wavelength of 3. We found no line emitter that is not associated

with a detected continuum source. However it remains possible that deeper exposures would lead to a secure detection of more redshifts from continuum features, possibly increasing the number

of groups or including fainter members. Fig.8shows the location

of the detected galaxies in groups within the MUDF footprint.

5.1 Going beyond the survey edges

The unprecedented depth of the MUSE observations in the MUDF comes at the price of a relatively small survey area. Therefore, it is possible that at least some of the groups identified in the previous section are part of a larger scale structure, or that some of the group members are missed by our footprint. Unfortunately, this area of the sky is not covered by wide-area archival spectroscopy, thus we can only attempt to characterize the large-scale environment around the MUDF with photometric redshifts (photo-z).

For this purpose, we downloaded source and magnitude

cata-logues from the DES DR1 (Abbott et al.2018), and we computed

photo-zs from the g, r, i, z fluxes using the EAZY code (Brammer,

van Dokkum & Coppi2008). Following Hoyle et al. (2018) we do

not include Y -band magnitudes, because the observations are too shallow to improve the photo-z estimate. These authors also studied the photo-z uncertainty on a smaller region of the DES footprint,

finding that they range from σz= 0.1 at z = 0.5 to σz= 0.4 at z =

1.5. For each group identified in the MUDF, we compute the galaxy

Figure 8. Gallery of the location within the MUDF of the galaxies in groups, marked with ellipses as in Fig.4. The white crosses mark the position of the background quasars, while the red crosses in each panel indicate the geometrical centre of the group. The cyan cross in the top left-hand panel marks the position of a serendipitously detected quasar that shows absorption at the redshift of this group (see Section 6.2). The number next to each galaxy shows the velocity offset with respect to the redshift of the group. The red segment in the bottom right-hand corner of each panel marks a 100 kpc scale at the redshift of each group.

density in an aperture with radius 1 Mpc centred on the geometrical centre of the group. The density is defined as the sum over all the galaxies within this aperture of the fraction of the DES photo-z

probability density function that falls within v= ±5000km s−1of

the redshift of the group. This approach optimally takes into account the variable photo-z accuracy as a function of redshift and galaxy

magnitude (Kovaˇc et al.2010; Cucciati et al.2014). We then define

two control apertures of the same size and velocity depth bracketing in redshift of the previous aperture, but not overlapping with it.

Despite the limitations of this approach, we do not find a significant overdensity of bright galaxies over a scale larger than the

MUDF footprint for any of the groups studied. We therefore con-clude that our inferred group properties are broadly representative of the underlying population. Nevertheless, wider area multi-object spectroscopic observations are required to investigate with better precision the large-scale environment of the MUDF.

6 T H E C G M O F G R O U P G A L A X I E S

Having completed a deep census of galaxy groups including

mem-bers that are as faint as mMUSE≈ 28.5 mag (or stellar masses down to

≈ 108_M

∗at z= 1.5), we now take advantage of the presence of the

two bright quasars in the field to probe the gas content of galaxies within these structures in absorption. In particular, we focus on the

MgIIabsorbers detected in the quasar spectra at z < 1.5.

6.1 MgIIfitting in the high-resolution quasar spectra

To identify and fit the MgII absorbers in the quasar spectra we

employed a two step procedure. First, strong absorption lines were searched for by visually inspecting the spectra across the entire

wavelength range, looking first for the MgIIdoublets and then for

other transitions which are commonly detected in quasar spectra,

e.g. MgI, FeII, MnII, CrII, and CaII. This procedure has been

repeated by two authors independently (MFo, VDO) and for both quasars. We reliably identified five absorbers in the bright quasar at

z≈ 0.67, 0.88, 0.98, 1.05, 1.15, while in the fainter quasar we only

identify one absorber at z∼ 0.88.

For each identified absorber we then fit the MgIIdoublet profile

with a novel method that uses Bayesian statistics to identify the minimum number of Voigt components required to model the data. Our method has two desirable properties: first, it requires minimal input from the user as no initial guesses are required; secondly, the final result is an optimal statistical description of the data, as the fit does not depend on any particular choice of initial guesses nor on a user-defined prior on the number of components. The details of the algorithm used will be presented in a forthcoming publication, and are only briefly summarized here.

Once the atomic transition (or transitions in case of multiplets to be fit jointly) has been identified, it is only required to specify the wavelength range (or disjoint ranges) to fit the normalized spectrum. The code then performs a fit of the data with progressively higher number of Voigt components (between a minimum and a maximum number), where each component is defined by a column density (N),

a Doppler parameter (b), and a redshift (z). For our fitting of MgII

absorbers we define uniform priors on the first two quantities in the

range 11.5 < log (N/cm2_{) < 16.0 and 1 < b/km s}−1_{<30, chosen to}

avoid modelling features within the noise or large-scale variations in the continuum level. The prior range on redshift is instead internally defined such that the absorption components can cover the selected wavelength range. A constant continuum level is also left as a free parameter in the fit to account for slight imperfections in the normalization of the spectra. If required following a visual assessment of the doublet profiles, we allow for a user defined number of ‘filler’ Voigt profiles designed to describe absorption lines arising from blends of different ions at different redshifts. One

example is the rightmost line in the profile of the z≈ 0.98 absorber

in Fig.9. The model spectrum is then convolved to match the line

spread function of the observed data using a Gaussian kernel. At each iteration, having defined a model with n components, we

sample the multidimensional likelihood space usingPOLYCHORD

(Handley, Hobson & Lasenby2015), a nested sampling algorithm

that has better performance thanMULTINESTfor high-dimensional

parameter spaces with multiple degeneracies between parameters. This is indeed our case, where complex models can reach a number of free parameters in excess of 50. After the sampling, we use the posterior samples to derive the Akaike Information Criterion (AIC;

Akaike1974) of each fit, defined as AIC=2 × npar− 2 × ln(L),

where nparis the number of free parameters and L is the likelihood

of the model. We select as optimal fits those with a likelihood ratio within 1:150 compared to the best (lowest) AIC. If the selection includes only one model this is used as our best fit, otherwise we choose (from the selected ones) the model with the minimum number of components.

Fig.9shows the fits of the MgIIabsorption systems identified

in the quasar spectra. While the fits are jointly obtained from the 2796 and 2803 Å transitions, we show only the stronger 2796 Å transition in this figure. The single absorber found in the fainter quasar, which we discuss in more detail in Section 6.3, is shown in the Bottom Right panel, while in the other panels we show the absorbers found in the bright quasar sightline. In all panels we find complex kinematic profiles requiring from 3 to 15 Voigt components. Moreover we note that, with the exception of only one

system (z∼ 0.98), the absorbers are found at the redshift of a galaxy

group identified in the previous section. Therefore, we set the zero of the velocity axis at the redshift of the galaxy group, except for one absorber where it is set to the redshift of the closest galaxy (ID 14) in velocity space. We stress that the position of the zero on the velocity axes is only made for reference and it does not affect our results.

6.2 Connecting the absorption features and the galaxy population

The fact that MgIIabsorbers are preferentially found at the redshift

of a galaxy group (5/6 cases) suggests a direct connection between overdensities of galaxies and large cross sections of cold gas. Leveraging the deep spectroscopic survey in the MUDF that is complete to faint levels/low masses, we systematically explore this connection, in order to understand the physical origin of the gas probed in absorption.

We show in Fig.9the spectroscopic redshift of the galaxies in

the MUDF which are within the plotted range for each absorber (blue stars). We immediately note two features: first, each absorber

(with the exception of the one at z≈ 0.98) has one or more galaxies

which are overlapping in velocity space with the absorption profile.

Secondly, the absorbers with more complex kinematics (z∼ 0.88,

and z∼ 1.05) have the largest number of galaxies at a small velocity

offset. The case of the z∼ 0.98 absorber, instead, stands out for

having a complex profile, yet a single galaxy which is offset in velocity. However, given that the bright quasar is close to the South– East edge of the observed field, we cannot rule out the presence of other objects, just outside the MUDF footprint but at smaller velocity separation compared to the galaxy we found. Another issue is that the detection of continuum sources is more difficult at small projected separation from the quasars due to their brightness. We

can detect sources as faint as mMUSE 27 mag at ≈25 and ≈18 kpc

in projection from QSO-SE and QSO-NW, respectively. However, brighter or highly star forming galaxies would be detected even at smaller projected distances.

Thanks to the integral-field spectroscopic coverage of the MUDF, we next investigate trends in the spatial location of the galaxies which are found in the velocity window defined by the absorbers.

Starting with the z∼ 0.67 absorber as a first example, we note how

the MgIIprofile spans a velocity range v∼ 100−200 km s−1with

respect to the redshift of the group. We recall that the latter quantity is the average of the redshift of the group galaxies, and so we should expect a roughly equal number of them above and below the average value. However, not all of them need to be distributed isotropically

with respect to the quasars. In fact, by looking at Fig.8, we find

that the four galaxies with velocities overlapping with the MgII

absorber are typically closer to the bright quasar than the remaining galaxies in the group. Indeed, their average projected distance from

QSO-SE is ≈130 kpc, while for the other galaxies it is twice as

large, at≈250 kpc.

Despite the large uncertainties driven by small number statistics, a similar trend emerges from the analysis of the other absorbers

Figure 9. UVES spectra of MgIIabsorbers identified in the brighter QSO-SE (all panels except last two rows in the rightmost column), and in the fainter QSO-NW (last two rows in the rightmost column). The zero velocity corresponds to the average redshift of the galaxy group associated with the absorber, or to the redshift of the closest galaxy for the z≈ 0.98 absorber. The observed spectrum is shown as a black histogram with the 1σ error array in grey. A Voigt profile Bayesian fitting gives the red models, where individual lines are sampled from the posterior distribution. We include the rest-frame total equivalent width measurements in each panel. The red ticks above the continuum level show the position of individual MgIIabsorption components, which are shown as cyan dotted lines. The pink dotted lines are for filler components arising from blends of different transitions. The blue stars show the spectroscopic redshift of MUDF galaxies.

Table 4. Average projected distance of the galaxies which reside in groups at the same redshift of a MgIIabsorber, for galaxies within the velocity window defined by the absorption profile, and outside it. For groups without an MgIIdetection the average projected distance of all galaxies is given in the last column. Values without errors have only one object that satisfies the selection threshold.

zabs Group ID din, win(kpc) dout, win(kpc)

0.67 1 130± 56 251± 25 0.68 2 – 257± 94 0.78 3 – 363± 92 0.88 4 274± 87 522 1.05 5 162± 41 467± 29 1.15 6 195 321± 101 1.22 7 – 322± 83

associated with the groups, as summarized in Table4. This table

re-ports the typical distance from the sightline QSO-SE of galaxies that

overlap in velocity with the absorption profile (din, win), compared

to the distance for the remaining galaxies in the group. In all cases, galaxies overlapping in velocity with the absorption profile lie at

closer projected separation (by a factor of≈1.5−2) than galaxies

that are not overlapping with the MgIIabsorbers.

Considering instead the groups with no MgII association (ID

2, 3, and 7), Fig.8 reveals a lack of galaxies in the immediate

surroundings of the QSO-SE line of sight, in line with the trend above. These results therefore suggest that the cold gas absorbers

traced by MgIIarise preferentially from the CGM of one or multiple

galaxies which are closer than≈100 kpc from the quasar line of

sight. It should however be noted that the probability of group

galaxies giving rise to MgIIis unlikely to be exclusively modulated

by proximity to the line of sight, with covering factors playing a

role. Indeed, Fig. 8shows also the presence of galaxies in close

proximity to the QSO-NW line of sight (mostly from groups 3, 4, and 5) that do not necessarily give rise to absorption (see below).

The relation between the strength of MgII absorbers and the

galaxy distance has been studied extensively in the literature. For

instance, Chen et al. (2010) report a large statistical sample of

galaxy–quasar pairs, finding a strong anticorrelation between the equivalent width of the absorption profile and the distance of the closest galaxy. We can therefore compare directly in this parameter space the properties of MUDF galaxies, both within groups and in isolation, with the results present in the literature.

In Fig.10, we show the rest-frame equivalent width (W) for the

MgIIλ2796 Å absorption line, obtained integrating the best-fitting

models derived above, as a function of the projected distance of

Figure 10. Rest-frame equivalent width (W) for the MgIIλ2796 Å absorption line as a function of impact parameter (projected distance of each galaxy from the QSO-SE sightline for MUDF galaxies, left-hand panel) and of the galaxy stellar mass (right-hand panel). Galaxies in the MUDF footprint are shown as red circles if they reside in a group and as blue circles if they are isolated. The arrows (red for galaxies in groups and blue for isolated galaxies) show 2σ upper limits measured on the UVES spectrum in a 10 km s−1velocity window centred at the redshift of each galaxy. The black empty stars show galaxies which have a redshift within the MgIIabsorption profile for the strong absorbers shown in Fig.9. The redshift of the groups are shown in the left-hand panel at the y-axis levels of the corresponding values of W. The black line is the best-fitting relation (with the 1σ confidence area shaded in grey) between W and impact parameter from Chen et al. (2010). The orange stars and black circles are obtained from stacking galaxy–galaxy pairs from the sample of isolated galaxies of Rubin et al. (2018a) and the sample of group galaxies of Bordoloi et al. (2011).

each galaxy from the QSO-SE sightline (left-hand panel) and of its stellar mass (right-hand panel). Galaxies belonging to a group that

is associated with an MgIIabsorber are shown as red circles and

they have been assigned the total W of the absorption profile, while

the galaxy which we associate with the z∼ 0.98 absorption system

is shown as a blue circle. For the other sources, we plot 2σ upper

limits measured on the UVES spectrum in a 10 km s−1velocity

window centred at the redshift of the galaxy.

From this analysis, it appears that the four out of seven groups that are associated with an absorber have high equivalent width and, as noted above, have at least one galaxy which is within 150 kpc of the quasar line of sight. Moreover, these groups tend to host galaxies

that are relatively massive (log(M_∗/M_) > 10) and therefore, given

their CGM is likely to scale with their size (e.g. Chen et al.2010),

they are more likely to have a larger cross-section of MgIIthat can

give rise to the absorption.

An apparent exception is the z≈ 0.67 group, which seems to lack

particularly massive galaxies. However, we argue that in this case

the MgIIabsorption is mostly driven by the fact that one galaxy,

despite being of a lower stellar mass (log(M∗/M)= 8.92), is

very close in projection (29 kpc) and its redshift places it at the centre of the absorption profile. On the other hand, for the three groups without a detection, their galaxies are both further away

from the absorber and their stellar masses are low log (M_∗/M_)∼

9, overall reducing the chances that the CGM of group members intercepts the line of sight.

This picture is further reinforced when we make the same analysis for the QSO-NW. We find that the galaxies in groups are on average further away from this line of sight compared to that of the brighter quasar. Only three galaxies are within 100 kpc, two of which are in

the z≈ 0.88 group which we associate with the only MgIIdetection

in this sightline (see Section 6.3). Moreover, a third lower redshift

quasar (z≈ 1.285; cyan cross in Fig.8), lying close to the edge of

the FoV (where no sources have been extracted) is in close spatial

proximity to the z≈0.67 group. It does in fact exhibit strong MgII

absorption at this redshift as seen from the MUSE spectrum, but we did not find other absorbers associated with groups or individual galaxies at larger separations from this third sightline. We therefore conclude that a combination of proximity to the quasar line of sight and presence of massive galaxies hosting large cross sections of cool gas are the main factors that control the high incidence of

MgIIabsorbers in these groups.

Having established what drives the incidence of MgIIin these

groups, we now compare the absorption properties to samples of

galaxies from the literature. To this end, in Fig.10, we show the

results of the analysis by Chen et al. (2010) (black solid line) and the

average properties of the CGM of galaxies at 0.35 < z < 0.8 from foreground and background galaxy pairs in the PRIMUS survey

(Rubin et al.2018a). These galaxies are not explicitly selected to

be within groups. These authors find that the strength of MgII

absorption declines as a function of impact parameter (d, i.e. the projected separation of a galaxy from the probing sightline), with the average gas content of star forming galaxies versus d in the PRIMUS survey being consistent with the fitting function of Chen et al.

(2010), albeit with a large intrinsic scatter for individual galaxies.

However, the total equivalent width of MgIIin galaxies within the

MUDF groups is higher than the 1σ scatter of the data from Rubin

et al. (2018a) when shown as a function of d, but are consistent with

these data as a function of stellar mass. Likewise, compared to the

scaling relation of Chen et al. (2010), the MUDF group galaxies

lie consistently above the relation at fixed impact parameter. These results imply that the group environment has an effect on the

cross-section of MgII, leading to enhanced equivalent widths at larger

distances compared to galaxies not explicitly selected within groups.

Bordoloi et al. (2011) presented the radial MgIIabsorption profile

from a sample of group galaxies from the zCOSMOS survey finding that groups have more extended absorption profiles compared to a sample of isolated galaxies from the same survey. However, the radial profile for group galaxies is largely consistent with the one

from the Chen et al. (2010) and Rubin et al. (2018a) samples. This